With the support of several successful randomized placebo-controlled trials, the FDA approved Truvada for daily oral pre-exposure prophylaxis of HIV (PrEP) in the United States in 2012, and shortly thereafter the CDC and WHO released guidelines for widespread PrEP use by all at-risk individuals around the world. However, PrEP rollout is still is infancy, and there are several important questions regarding PrEP implementation that cannot be addressed by randomized trials. The causal transportability theory developed by Pearl and Bareinboim is a mathematically-grounded framework used to consider how effects observed in one setting might be applied to another. This dissertation proposes novel ways transportability can be applied to improve how trial results are used to inform implementation of PrEP. applies transportability to address some of these lingering questions about PrEP implementation.
The first chapter uses transportability to better understand why randomization to PrEP was effective in preventing HIV among cisgender men but not effective among transgender women in the iPrEx study. We find that after transporting the results of the trial from cisgender men to transgender women, differences in measured baseline characteristics between the populations were sufficient to explain the observed effect heterogeneity in the trial. The second chapter demonstrates how transportability can be applied to subgroup analyses of randomized controlled trials to produce target-specific guidance for how to most efficiently implement new interventions. To illustrate this approach, we transport subgroup analyses of the iPrEx trial to two hypothetical target populations and show that the subgroups with the lowest number needed to treat differs depending on the composition of the target population. The third and final chapter addresses a common practical challenge faced in applying transportability theory to real-world data: how to decide which variables to include in a transport estimator. In this chapter, we discuss the various types of unnecessary variables that may inadvertently be included in transport estimators. We use a Monte Carlo simulation study to identify what types of variables should be included to maximize the performance (with respect to mean-squared error) of the parametric g-computation transport estimator.
Together these projects highlight how transportability theory can be applied to improve translation of study results to real-world populations.
